An Anthropomorphic perspective on the Automated Facial Recognition System: Government’s Black Mirror Project
The Central government recently announced that it has given its approval for the implementation of the Automated Facial Recognition System (AFRS).[1] This system, once implemented across the country, will be able to, inter alia, extract an image from a video and match it with the image of an individual whose record is already in an existing database.
This article does not posit the legal or ethical implications of the employment of such technology and instead deconstructs what is going on behind this technology to encourage readers to think critically about them as this is a technology that will have broad social impacts which is potentially dangerous to health & well-being.
Through-out history, representations have both excited us and given pause or even caused mass panic.[2] What is different with the technological evolution with our current situation is that the unintended consequences are not limited to individual responses that pass but rather are capable of systematic effects that we have difficulty seeing and that could cause great harm.[3] The trouble arises from the sociological fact that there has been a recession of reality as the culture has been absorbed into an increasingly ubiquitous, technological space with its own utopian visions.
One such artifact of the current times is the Automated Facial Recognition System that involves the exercise of power by inscrutable algorithmic authority and bears the potential to exponentially recede our fabric of reality.
Automated Facial Recognition System employs the differing instantiations of computing vision where machine vision substitutes human vision in technological initiatives that will have broad social impacts.
THE POWER OF PERSPECTIVE — A METAPHOR FOR UNDERSTANDING CONTEMPORARY VISUAL TECHNOLOGIES
While delving into the power of perspective in Renaissance painting, perceptual psychologist Michael Kubovy laments that “we must perceive the window in order to see the world.”[4] It means that in order to truly see the world we must see our own point of view in the perception of it. What about a scenario where we look not at the world but increasingly at the world re-presented through mediation? Our devices with their black mirror screens, when running, show representations of content, not the things themselves. This is inherently problematic because:-
(1) Humans mostly just take the picture as given and don’t notice the mediation — this is naïve realism ;
(2) The Power of the frame shapes pictorial dynamics & edges put pressure on visual elements subtly altering the relative importance in the field.
Pictures are spatial constructs. Spatial construction allows very dissimilar items to sit together without yielding their places. In this regard, the screen on which we view content continues and grows out of embedded cultural habits.[5] While viewing any picture what attracts our eyes first primes our reception of it and impacts the meaning we attribute to the representation as a whole. In contrast to the written language that follow the rules of their grammar(s), pictures are held together by association within the defined space and the spatial relationships that occur there.
Now, with the ingress of artificial intelligence, the display devices have led to the promulgation of something called the “frame problem”. The frame problem has to do with the “difficulty of defining what information or inferences may be relevant or irrelevant to a problem.” The availability of perceptual frame says “these things are relevant to your attention now.” If there are multiple pieces of content within the largest frame and many sub-windows, the viewer must still determine what is relevant to his or her interests, a far more complex process than viewing a single picture in a frame, even if it contains various modes of representations as in the examples above.
Display content come in framed units — windows, tabs, menus or messages but however many there are, they hang within an infinitely flexible space where nothing actually touches the edge, decreasing the awareness of the frame but not it’s power.[6]
AUTOMATED FACIAL RECOGNITION SYSTEMS — NOT A FOOL-PROOF SYSTEM
AFRS is a variant of computer vision that is task specific. It is tailored to seeing something in particular and limited by that; human eyes, in contrast, have specific sensors that can perceive, but not necessarily understand, any kind of thing that is visible to them. Computer vision relies on data from sensors (in AFRS, pictures in a database that may derive from different forms of recording that may have been regularized for position or other considerations before being included), and an algorithm (software procedural instructions) that has to be trained to perform matching tasks. It must be remembered that intelligence is limited to its task. Training is done by humans and relies on Datasets where each relevant element is identified (labelled) by humans, at least in the beginning. However, relevant elements do not account for everything in the picture. Labelling involves human selection of relevant information and giving it a name, [7] presumably from predetermined categories reflected in the configuration of the database holding the dataset. Interestingly, that is no guarantee from the freedom of bias.[8]
Now, the idea of being able to uniquely identify individuals is very appealing, and a 90% accuracy rate might seem to be very good, looked at from the point of view of a very large system. However, if you are one of ten in a hundred who is inaccurately identified, you might find this system a problem.
The introduction of efficiency enables control of a kind not fathomable otherwise. Nevertheless, is it good for human freedom if everyone is in The Perceptual Line-up?[9] Does the dominant use of technology for law enforcement and security not cast a shade of guilt on anyone whose face appears in it even if they were accidentally in a scene depicted? How legitimate is it for an entity as large as Amazon, with accumulated personal data on its vast customer base, to provide its information service Rekognition,[10] when its customers have no idea what information about them could be shared and whether it has been combined with another set of data?
AFRS: VISUAL TOOL FOR SOCIAL CONTROL & POTENTIAL BLACK MIRROR
It is not that visual technologies have never been used for purposes of social control. Let’s not frame it as a conversation about balancing a system that moves past human control and can be put to bad purposes — rounding up members of the population whom the government or other powerful forces wish to disadvantage or put under surveillance; to assist in denying opportunities for any reasons thought useful by those with power in the system. Instead, let’s try to bunk the notion the rhetoric about AI that it is more intelligent than human beings. The software can certainly process more data from databases more quickly than we humans can. However, the algorithm,a set of defined procedures cannot tell us how it made its decision, because while its procedures are specified by the code,its judgments and their reasons are hidden from us because they are the product of the machine’s own learning. They are not conscious; they cannot explain their criteria;they cannot be argued with. And they are only as good as the data they are given. So while the apparent concern stemming out seems to be is how the probe picture is understood by the AFRS (in or not in its database); howsoever this is just a surface concern and in fact the deeper concern lies in the process of the algorithm.
To put in a nutshell, the AFRS requires information to be abstracted, processed and then finally the software is implemented for specific tasks through training which may then refine it further using machine learning, a kind of software. Now those who rent or buy the tools may not have any idea how they work or what their implications are.[11] It is important to be kept in mind that those AFRS databases involve pictures AND keywords, the image and all the metadata associated with it. Pictures are never simple[12] and “Language is not transparent”. Just because a face expresses affect does not mean that we can easily know, or guess, the meaning behind the expression or emotion in another, what its subjective meaning is and what intention it might betray.
The ubiquity of photography everywhere in our lives accommodates us to its essential flatness and yet it is the three dimensionality of real people that help us to identify them in life. The more naturalized the tool, I say, the less we are able to think critically about its assumptions. AFRS is such a tool in a black mirror that wears algorithmic masks for photographic actions.
REFERENCES :-
[1] See https://pib.gov.in/newsite/PrintRelease.aspx?relid=199808.
[2] Boyd Brian, On the origin of stories: Evolution, cognition, and fiction. Cambridge. MA Belknap Press of Harvard University Press. (2009).
[3] Ibid.
[4] Michael Kubovy, The psychology of perspective and renaissance art. Cambridge University Press. (1986).
[5] See E.H. Gombrich, Standards of Truth: The Arrested Image and the Moving Eye Critical Inquiry, 7, no. 2 237–273 (1980).
[6] Interview with Daniel Kiecza, Software Engineer, Google, Inc. (Jan 18,2014).
[7] White paper by Figure Eight, What We Learned Labeling Million images, A Practical Guide To Image Annotation For Computer Vision ln.d.
[8] Buoamwini, Joyland Gebru, Timnit, Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification, Proceedings of Machine Learning Research 81 (2018); See also https://www.independent.co.uk/news/uk/home-news/met-police-facial-recognition-success-south-wales-trial-home-ofce-false-positive-a8345036.html; Patric Grother, Mei Ngan, Kayee Hanaoka, Face Recognition Vendor Test (FRVT) Part 3: Demographic Effects. 2019. available at https:// doi.org/10.6028/NIST.IR.8280.
[9] See Garvie, Clare et.al. The Perpetual Line-up, Unregulated Police Face Recognition in America, October 18, 2016.https://www.perpetuallineup.org/.
[10] Kaste, Martin “Orlando Police Testing Amazon’s Real-Time Facial Recognition,” May 22, 201810:01AM ET. Audio on All Things Considered, https://www.npr.org/2018/05/22/613115969/orlando-police-testing-amazons-real-time-facial recognition. Al 2019 figure for American Customers of Amazon is 103 million. https://www.oberlo.com/blog/amazon-statistics.
[11] See Citron, Danielle Keats, Technological Due Process, Washington University Law Review, 85, 6, (2008).
[12] Spiesel, Christina , Reflections on Reading: Words and Pictures and Law. 2009.
